During this time, the exposure to nVIDIA, AMD’s next-generation flagship GPU core continues, but all kinds of rumors, elusive. Now, we’ve dug out two new cores of NVIDIA from the GeekBench 5 test database, both of which are quite large. One shows 118 computing units (NVIDIA is called SM streaming multiprocessor), if each group continues 64 stream processors (CUDA core) configuration, the total is 7552, while the core frequency is 1.11GHz, with a memory of 24GB (detection value 23.8GB).
The other is 108 computing units, corresponding to 6912 stream processors, with a core frequency of 1.01GHz, with a memory of 48GB (detection value of 46.8GB).
Obviously, they are the next-generation core of NVIDIA, perhaps the fabled “Ampere”, 7nm process, and possibly even the next generation of “Hopper”.
NVIDIA has also done dual-architecture core parallel series, such as the Tesla V100 series is the Volta core, for the professional market only, GeForce RTX 20, GTX 16 series is turing core , only in the consumer market.
In the absence of other corroboration, it is not possible to confirm which of the two cores belong to the computing professional card Tesla, the graphics professional card Quadro, the semi-professional card Titan, and the game card GeForce, and that Tesla is likely to be developed and released according to NVIDIA’s product development and distribution rules. The Tesla A100, for example, does nuns, but does not rule out The GeForce, such as the RTX 3080 Ti.
If the large core of 118 units and 7552 stream processors were THE RTX 3080 Ti, it would be 74% larger than the current RTX 2080 Ti of 68 units and 4352 stream processors.
Moreover, it is likely that this is not yet the full version of the core, 128 units is the most reasonable full strength, that is 8192 stream processors, compared to the RTX 2080 Ti is 88% larger than the 72 units 4608 stream processors, the full TU102 core Titan RTX is also 78% larger.
24GB memory if the game card GDDR6, compared to the RTX 2080 Ti can be more than doubled, and if it is a professional card HBM2, compared to the current Tesla V100 16/32GB is only a low version.
The core of 108 units and 6912 stream processors is certainly not RTX 3080, which will increase by 135% over the size of THE RTX 2080, which is outrageous, and the memory capacity of up to 48GB is impossible to use for game cards, but only For The Tesla or even quadro professional cards.
As for the low frequencies of 1.01GHz and 1.11GHz, it is easy to understand that they are all early engineering samples, which appeared at the end of October and the end of November last year.
While the core size has skyrocketed, even at such low frequencies, the 118-unit, 108-unit OpenCL computing runs are 184096, 141654, as reference tesla V100 154606, QuadroGV100 142837, Titan RTX 132804. Tesla V100
Regardless of the identity of these two cores, AMD will face a huge test, especially in the case of game cards, the legendary RDNA2 architecture, Big Navi’s Big Core, next-generation game cards are likely to have only 80 sets of computing units and 5120 streaming processors.
The next generation of the calculation card, the Radeon Instinct MI100, codenamed Arcturus, is said to double on the Vega core, with 128 computing units, 8,196 stream processors, and power-controlled at 200W in the region.